Evaluating the Impact of Alternative Dependency Graph Encodings on Solving Event Extraction Tasks

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 982–992, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Evaluating the Impact of Alternative Dependency Graph Encodings on Solving Event Extraction Tasks Ekaterina Buyko and Udo Hahn Jena University Language & Information Engineering (JULIE) Lab Friedrich-Schiller-Universit¨at Jena F¨urstengraben 30, 07743 Jena, Germany {ekaterina.buyko|udo.hahn}@uni-jena.de Abstract In state-of-the-art approaches to information extraction (IE), dependency graphs constitute the fundamental data structure for syntactic structuring and subsequent knowledge elicita- tion from natural language documents. The top-performing systems in the BioNLP 2009 Shared Task on Event Extraction all shared the idea to use dependency structures generated by a variety of parsers — either directly or in some converted manner — and optionally modified their output to fit the special needs of IE. As there are systematic differences be- tween various dependency representations be- ing used in this competition, we scrutinize on different encoding styles for dependency in- formation and their possible impact on solv- ing several IE tasks. After assessing more or less established dependency representations such as the Stanford and CoNLL-X dependen- cies, we will then focus on trimming opera- tions that pave the way to more effective IE. Our evaluation study covers data from a num- ber of constituency- and dependency-based parsers and provides experimental evidence which dependency representations are partic- ularly beneficial for the event extraction task. Based on empirical findings from our study we were able to achieve the performance of 57.2% F-score on the development data set of the BioNLP Shared Task 2009. 1 Introduction Relation and event extraction are among the most demanding semantics-oriented NLP challenge tasks (both in the newspaper domain such as for ACE1, as well as in the biological domain such as for BioCre- ative2 or the BioNLP Shared Task3), comparable in terms of analytical complexity with recent efforts di- rected at opinion mining (e.g., NTCIR-74 or TREC Blog tracks5) or the recognition of textual entail- ment.6 The most recent BioNLP 2009 Shared Task on Event Extraction (Kim et al., 2009) required, for a sample of 260 MEDLINE abstracts, to determine all mentioned events — to be chosen from a given set of nine event types, including “Localization”, “Binding”, “Gene Expression”, “Transcription”, “Protein Catabolism”, “Phosphorylation”, “Posi- tive Regulation”, “Negative Regulation”, and (un- specified) “Regulation” — and link them appropri- ately with a priori supplied protein annotations. The demands on text analytics to deal with the complex- ity of this Shared Task in terms of relation diversity and specificity are unmatched by former challenges. For relation extraction in the biomedical domain (the focus of our work), a stunning convergence towards dependency-based syntactic representation structures is witnessed by the performance results of the top-performing systems in the BioNLP’09 1http://papers.ldc.upenn.edu/LREC2004/ ACE.pdf 2http://biocreative.sourceforge.net/ 3www-tsujii.is.s.u-tokyo.ac.jp/GENIA/ SharedTask/ 4http://research.nii.ac.jp/ntcir/ workshop/OnlineProceedings7/pdf/revise/ 01-NTCIR-OV-MOAT-SekiY-revised-20081216. pdf 5http://trec.nist.gov/data/blog08.html 6http://pascallin.ecs.soton.ac.uk/ Challenges/RTE/ 982 Shared Task on Event Extraction.7 Regarding the fact that dependency representations were always viewed as a vehicle to represent fundamental seman- tic relationships already at the syntactic level, this is not …

Content truncated to 500 words